Re: [Xen-devel] [PATCH] xen/arm: gic-v3: Implement correctly the callback send_SGI

[Julien Grall <julien.grall@xxxxxxxxxx>]

Hi Ian,

On 08/05/15 15:01, Ian Campbell wrote:
> On Mon, 2015-04-27 at 19:31 +0100, Julien Grall wrote:
>> Currently, the GICv3 drivers is only able to send an SGI when the cpumask
> 
> "driver"
> 
>> is provided. Although with the modes SGI_TARGET_OTHERS and SGI_TARGET_SELF,
>> no cpumask is provided. Any usage of those modes will crash the hypersivor.
>>
>> Move the rename gicv3_send_sgi to gicv3_send_sgi_list and implement the
> 
> s/Move the r/R/

I missed while reviewing the commit message. I will fix it in the next
version.

>> different modes:
>>     - SGI_TARGET_OTHERS: Set the Interrupt Routing Mode (bit 40) to 1
>>     (see Table 4 on Section 4.2.6 PRD03-GENC-010745 24.0)
>>     - SGI_TARGET_SELF: Contrawise GICv2, the SGI registers doesn't provide
> 
> "Contrariwise"? But I think you really mean "Unlike". I'd also say "the
> GICv3 SGI registers" and s/doesn't/don't/.

Yes, "unlike" is better here.

> 
>>     a specific field. So use gicv3_send_sgi_list and pass the cpumask of
>>     the current CPU
>>     - SGI_TARGET_LIST: Directly call gicv3_send_sgi_list with the given
>>     cpumask
>>
>> Reported-by: Chen Baozi <baozich@xxxxxxxxx>
>> Signed-off-by: Julien Grall <julien.grall@xxxxxxxxxx>
> 
> Apart from the commit message nits one code comment:
> 
> [...]
>> +    case SGI_TARGET_OTHERS:
>> +        WRITE_SYSREG(ICH_SGI_TARGET_OTHERS << ICH_SGI_IRQMODE_SHIFT |
>> +                     (register_t)sgi << ICH_SGI_IRQ_SHIFT,
> 
> I don't think you want register_t here, unless ICC_SGI1R_EL1 might be a
> different width on different architectures.

Yes, ICC_SGIR1_EL1 is always 64-bit.

> I think you should just cast to uint64_t and WRITE_SYSREG64.

Right, I blindly re-use the WRITE_SYSREG as it was done on the previous
implementation.

Although, the gic-v3 code seems to always use {READ,WRITE}_SYSREG. This
seems wrong too. I will send a patch for replace the calls by the the
64-bit version.

Regards,

-- 
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel